stretching a pulse

Why wouldn't you just stretch the pulses with logic circuits? Your delay element would be a string of inverters.

I forgot to answer your other question. The inverse sinc is a filter where you feed it an impulse and you get a pulse out. It is the theoretical PCM reconstruction filter. Remember, sampled data theory presumes impulses.

In real life of course, your DAC creates steps rather than impulses, so the inverse sinc is more theory than reality. You do need to boost the high frequencies of a staircase signal in your antialiasing/ reconstruction filter. They call it sinc compensation.

I've done that a couple of times now and it works fine. For one we tapered the coefficients slightly to compensate for the loss in the delay line, and for another I used 0.5, 1, ... 1, 0.5 to make the filter a little closer to Gaussian. Both worked adequately, but we didn't analyse them exhaustively.

The first time I had to tackle this particular problem, I used a monostable.

Ghiggino, K.P., Phillips, D., and Sloman, A.W. "Nanosecond pulse stretcher",Journal of Physics E: Scientific Instruments, 12, 686-687 (1979).

As a solution, it was pretty horrible - the threshold depended on the repetition rate, due to self-heating in the transistors, as is mentioned in the paper. An emitter-coupled monstable - with similar

5GHz transistors - should have done better, but the circuit was adequate in the application, where the repetition rate was a steady 20MHz (IIRR) set by the laser generating the pulses, and nobody wanted to fix something that wasn't actually broken.

An infinitely long FIR filter, for one.

-- Bill Sloman, Nijmegen

On a sunny day (Tue, 31 Aug 2010 21:29:17 -0700) it happened John Larkin wrote in :

One shot? There must be some logic that fast.

It has to be analog and linear. Downstream will be amplifiers and comparators, as noted.

Another possibility is two cascaded Bessel filters. The first shapes the 2 ns pulse into, say, a 6 ns gaussian pulse, and the second pretty much just delays that. The sum of the filter outputs will be pretty much flat and will settle out fast.

John

John

Can you slow it up at the source? Make the photodiode (PD) a bit slower? Are the pulses straight from a PD or is there some active element? Perhaps a bit more resistance somewhere.

George H.

A circuit made of lumped components can't have more zeroes then poles. Hence what you trying to build is a twisted elliptic lowpass filter where zeroes are at lower frequencies then the poles. I would approach this as an optimization problem.

Vladimir Vassilevsky DSP and Mixed Signal Design Consultant

We might try that. I can buy a 5-tap, 1 ns per tap, SIP delay line for about $5. There is some concern as to whether such parts will be available long-term, since the demand for analog delay lines seems to be fading.

Other possibilities: two cascaded 3 to 5-pole Bessel filters, maybe with a buffer between, and resistive summing of their outputs. The second one is pretty much acting as a delay.

A Bessel followed by all-pass sections, ditto.

John

IIRC if you use a constant-K filter, all sections except the ends have the same L and C values, so it looks like a transmission line. Tapping off from the intermediate nodes would work fine then.

I gather you care about the pulse height, and so can't just use (say) a TTL gate with a cap on its output. TTL's output drive is asymmetric enough that you should be able to stretch the pulse some that way.

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs
Principal
ElectroOptical Innovations
55 Orchard Rd
Briarcliff Manor NY 10510
845-480-2058
hobbs at electrooptical dot net
http://electrooptical.net

Yup, that would be a homemade lumped LC delay line. They are OK for modest lengths and modest delay/risetime ratios. They explode after that. The number of sections goes as Td/Tr squared, even if you can buy ideal inductors.

The PD pulse will drive, probably, four comparators, each programmed by a DAC. One will hunt for the nominal height, one will derive a trigger at maybe H/3, and two will look for over/under amplitude pulses. The amps, comparators, and FPGA would be happier with 5 ns or so pulses, but we need to recover pretty fast. Hence the rectangular output filter.

I can buy a 5-tap, 1 ns per tap SIP LC delay line for about $5, so maybe a bessel lowpass followed by a transversal filter would work.

Rob is trying to design a classic form LC filter that has a rectangular output, using a genetic-evolution optimizer thing. So far, the waveforms are, well, interesting. He's calling LT Spice to get the response, with that wrapped inside some error function + optimization thing, running for maybe 35K shots per trial. I'm thinking in terms of doing the design with a breadboard LC delay line and 4 or 5 trimpots. The cool thing about transversal filters is that they can be tuned by hand without difficulty; you practically draw the waveform you want.

John

nd

I claim pulse stretching is not a linear operation. I don't recall the strict definition of linearity, but it had something to do with energy conservation (Parseval). We're talking geek here, not stuff like low THD, etc. that is normally associated with linearity.

My guess is that you are trying to maintain the amplitude of the pulse. I can see that being designed with a peak detector and one shot.

nd

In the first example, I loaded each tap of a nominally 100R lumped constant delay line with a nominally 1k5 resistor running into a summing junction; by starting off with 1k7 and tapering down to 1k3 we more or less compensated for the pulse attentuation in the loaded delay line. The summing amplfier was something tolerably fast from Comlinear, though we might have switched to an Analog Devices part on the second spin of the board.

The second example was done some five years later, when you could get quad packaged op amps of similar speed, and I buffered each tap of the delay line with a follower amp and used the outputs of the followers to drive the resistors into the summing junction.

Apparently commercial delay lines use low-pass sections. The one time I got somebody to design a home-built delay line, comment was that that got you twice as much delay per unit R/C or L/C product. The guy who was doing the work was very good, but the project got canned before we got to the design review on that board (which really peeved me - we were designing a board to replace one that had been evolving for about ten years, and contained lots of evidence that evolution drives you towards sub-optimal solutions).

-- Bill Sloman, Nijmegen

The bad news is that few people still make analog delay lines, and they want 4 weeks to come up with a sample. I don't have 4 weeks. So either I make my own tapped LC delay line from parts, or try cascading and summing Bessel filters.

John

What I want to do certainly is. A passive LC filter is linear. The output is strictly proportional to the input.

John

OK, this is weird,

ftp://jjlarkin.lmi.net/Bessel_3+7c.jpg

pretty much just instinct and fiddling. It only uses 4 inductors, so it's not outrageous to actually do. It hardly has any tail, so I could slow down the filters a bit and stretch the output a little wider.

TAP is pretty far from F2, which means that most of the delay in the second filter happens in the last LC. Maybe I can flip ends on this filter and pretty things up a little. Or use something more like a classic delay line for the second section. Interesting, and mildly tedious.

T-coils make better delay lines, but it's hard to get the tapped inductors.

John

For 5 ns, you need about five feet of twinlead or coax cable to make a delay line. Probably there's a way to get it without a four week wait.

You can make a short pulse have a long tail by feeding it to a termination network with a diode, and the resulting (fast-rise, slow-fall) pulse can be shortened by using a delay line stub (the reflection is a fast-fall slow-rise pulse that, added to the long-tail original, has a fast-rise and fast- fall with nearly zero straggle).

d

an

Hz

ns

tty

t and

r

ate

Then you are not stretching the pulse. Think of it this way. The input pulse is of amplitude A and width W. The output pulse is of amplitude A and width Y, where Y>W. The energy of the output is greater than that of the input, hence my reference to Parseval.

Oh, eh yeah, QED. ;-)

Taps on a few feet of suitable coax?

Sylvia.

*Unless* you put it on narrower traces at the output (high Z). Then you get the same amplitude (voltage) but less power (since the energy is spread out) and a lot less current.

Tim

--
Deep Friar: a very philosophical monk.
Website: http://webpages.charter.net/dawill/tmoranwms